by Cactus
Compact 16.9 MB speech-to-text model from Cactus that runs on CPU with no dependencies.
Updated EU status checked Last shown
Sources: creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Whistle is a small speech-to-text (speech recognition) model from Cactus. According to a creator review, the model file is about 16.9 MB and runs on CPU with no external dependencies, which makes it suited to on-device and edge use.
It processes audio in 30-second windows, detects the spoken language automatically and returns word-level timestamps. The same review reports a lower word error rate and faster decoding than OpenAI's Whisper Base. These figures come from a creator's summary, not from an official page, and have not been verified here.
No official page, repository or model card could be fetched, so the licence, pricing, release date and distribution channels are unconfirmed.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
A 16.9 MB speech-to-text model that runs on CPU with no dependencies. It handles 30 seconds of audio, auto-detects language, gives word timestamps, and has lower error rate and faster decoding than Whisper Base.
Closest neighbours by meaning, not just by category.
Model · Edge0
Open-weights 0.17B-parameter multilingual text-to-speech model with zero-shot voice cloning, runs on CPU via ONNX.
Model · Nyra Labs
Controllable multilingual speech recognition with verbatim or clean output and word-level timestamps, from Nyra Labs.
Model · Cactus
Tiny on-device tool-calling and extraction model from Cactus, shipped as an 8–29 MB binary for phones, wearables and microcontrollers.
Model · Microsoft
Microsoft AI's low-latency text-to-speech model in the MAI model family, designed for long-form expressive speech.
Model · Google
Google's speech-to-text model with verbatim and smart modes, speaker detection, word timestamps and a live streaming variant.
Model · Cactus Compute
Tiny 29M–121M parameter on-device model for tool calling and extraction, shipped as an 8–29 MB binary.
